Submitted 29 September 2017Accepted 14 December 2017Published 10 January 2018

Corresponding authorRonald D. Hills, Jr, rhills@une.edu

Academic editorPedro Silva

Additional Information andDeclarations can be found onpage 17

DOI 10.7717/peerj.4230

Copyright2018 Hills, Jr

Distributed underCreative Commons CC-BY 4.0

OPEN ACCESS

Refining amino acid hydrophobicityfor dynamics simulation of membraneproteinsRonald D. Hills, JrDepartment of Pharmaceutical Sciences, College of Pharmacy, University of New England, Portland, ME,United States of America

ABSTRACTCoarse-grained (CG)models have been successful in simulating the chemical propertiesof lipid bilayers, but accurate treatment of membrane proteins and lipid-proteinmolecular interactions remains a challenge. The CgProt force field, original developedwith the multiscale coarse graining method, is assessed by comparing the potentialsof mean force for sidechain insertion in a DOPC bilayer to results reported foratomistic molecular dynamics simulations. Reassignment of select CG sidechain sitesfrom the apolar to polar site type was found to improve the attractive interfacialbehavior of tyrosine, phenylalanine and asparagine as well as charged lysine andarginine residues. The solvation energy at membrane depths of 0, 1.3 and 1.7 nmcorrelates with experimental partition coefficients in aqueous mixtures of cyclohexane,octanol and POPC, respectively, for sidechain analogs and Wimley-White peptides.These experimental values serve as important anchor points in choosing betweenalternate CG models based on their observed permeation profiles, particularly forArg, Lys and Gln residues where the all-atom OPLS solvation energy does not agreewell with experiment. Available partitioning data was also used to reparameterize therepresentation of the peptide backbone, which needed to be made less attractive for thebilayer hydrophobic core region. The newly developed force field, CgProt 2.4, correctlypredicts the global energy minimum in the potentials of mean force for insertionof the uncharged membrane-associated peptides LS3 and WALP23. CgProt will findapplication in studies of lipid-protein interactions and the conformational propertiesof diverse membrane protein systems.

Subjects Biophysics, Computational Biology, Computational ScienceKeywords Membrane protein, Molecular dynamics simulation, Hydrophobicity scale, Coarse-grained force field, Bilayer permeation

INTRODUCTIONCoarse-grained (CG) simulations greatly expand the timescale of events (Kmiecik etal., 2016; Venable et al., 2017) that can be studied in physiologically realistic membranesystems (Ma et al., 2015; Van Oosten & Harroun, 2016). Reducing the vast number ofatomic degrees of freedom in the many membrane lipids, large transmembrane proteinconstituents, and the surrounding bulk water is a necessary step that requires an accurateand transferable set of interaction potentials in the reduced conformational space (Bereau,Wang & Deserno, 2014). Construction of a standalone force field for the twenty amino

How to cite this article Hills, Jr (2018), Refining amino acid hydrophobicity for dynamics simulation of membrane proteins. PeerJ6:e4230; DOI 10.7717/peerj.4230

acids and common membrane lipids enables molecular dynamics (MD) simulation ofproteins of arbitrary sequence, structure and size (Bereau, Wang & Deserno, 2014; Ganesan& Matysiak, 2014; Han, Wan &Wu, 2008). By enabling membrane proteins to be studiedin the context of their native environment, CG simulations have yielded numerous insightsinto the nature and functional role of the dynamical interplay between proteins and thelipid bilayer (Bennett & Tieleman, 2013; Hedger & Sansom, 2016; Marrink & Tieleman,2013; Poyry & Vattulainen, 2016).

A natural method for assessing the balance of energetics in a CG model is to useumbrella sampling simulations to calculate the potential of mean force (PMF) for draggingsidechain analog compounds through the bilayer (MacCallum, Bennett & Tieleman, 2008).Agreement with the corresponding results from atomistic MD simulation has been used toassess a variety of CG representations such as Martini (De Jong et al., 2013), PRIMO (Karet al., 2014) HMMM (Pogorelov et al., 2014), ELBA (Genheden & Essex, 2015) and others(Vorobyov et al., 2016). The PMFs generated from atomistic MD of sidechains such as polarglutamine and basic arginine/lysine, however, have been observed to deviate from variousexperimental measures of solvation energy. Moreover, such studies neglect the influenceof both the polypeptide backbone and the occlusion of solvent-accessible surface areaby neighboring amino acid residues in a protein chain (Singh & Tieleman, 2011; Wimley,Creamer & White, 1996).

Experimental measures for the hydrophobicity of isolated sidechains involve thepartitioning of analog compounds (methane for Ala, propane for Val, etc.) in cyclohexane-water and octanol-water mixtures (Radzicka & Wolfenden, 1988). To estimate the solvationenergy for residues occluded by moderate-sized neighbor residues in the peptide chain,the octanol-to-water partitioning of Ac-WL-X-LL pentapeptides was determined (Wimley,Creamer & White, 1996) relative to AcWL-G-LL (see 1Gcor in Table 2 of Wimley et al.).Combining these values with the estimated 1.15 kcal/mol backbone penalty for nonpolarsolvation of the glycyl -CH2CONH- unit yielded a whole-residue hydrophobicity scalethat is useful for predicting transmembrane regions in proteins (White & Wimley, 1998).For a measure of exposed sidechains, solvation energies were corrected for occlusion bythe host peptide using measurements of the nonpolar solvent-accessible surface area ascompared to Ac-GG-X-GG (see 1GGXG in Table 2 of Wimley et al.). Another measurefor the fully exposed residue is the octanol-water solvation energy of an acetyl amino acidamide dipeptide: Ac-X-amide (Fauchere & Pliska, 1983). Experimental values are compiledin the Supplemental Information (Table S1, CSV format).

In the present work, parameter modifications of the CgProt force field (Hills Jr, Lu& Voth, 2010; Ward, Guvench & Hills Jr, 2012) are explored by comparing their resultingsidechain permeation profiles in a DOPC bilayer. By reassigning the CG site type definitionsfor select sidechain interaction centers, the permeation profiles of several amino acids arefound to improve relative to either atomistic simulation results or the known experimentalsolvation energies. The sidechain parameters are then coupled to an improved description ofthe peptide backbone. The resulting force field, CgProt version 2.4, is assessed for the abilityof designed helical peptides to remain inserted in the membrane during microsecond MDsimulations. While the previous version, CgProt 2.3, relied on terminal charges to anchor

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 2/22

transmembrane peptides across the bilayer, the current force field captures the properorientation of neutral peptides LS3 andWALP23 in the bilayer, consistent with experimentsusing capped uncharged termini. The refinement of amino acid hydrophobicity makesCgProt a powerful tool for studying lipid-protein interactions.

METHODSCoarse grain modelThe first iteration of the CgProt force field was developed for aqueous proteins usingthe multiscale coarse graining (MS-CG) method (Hills Jr, Lu & Voth, 2010). MS-CG is avariational force-matching procedure for developing a self-consistent set of CG interactionpotentials given a reference ensemble generated from atomistic MD (Noid et al., 2008). Astrength of MS-CG is that it can map an inherently many-body PMF into effective pairwiseinteractions without an a priori assumed functional form (Noid et al., 2007). To constructa protein force field, tabulated nonbond potentials were developed for five unique CGsite types ca/p/ap/pos/neg assigned to groups of backbone and sidechain atoms based onatomistic reference simulations of peptide unfolding in water. Because the model detailemphasized the amino acid sidechains rather than backbone, CgProt itself is not useful forde novo folding of proteins. When deployed with an elastic network encoding the backbonenative conformation, the force field can effectively simulate protein conformationaltransitions (Ward, Guvench & Hills Jr, 2012).

To enable the simulation of membrane proteins, atomistic simulations wereperformed of peptide unfolding in aqueous mixtures of phospholipids. MS-CG wasthen used to develop lipid-protein interaction potentials for the headgroup and tail sites:ch/hh/ph/gl/e1/e2/s1/sd/sm for common physiological lipids (Hills Jr & McGlinchey, 2016;Ward, Guvench & Hills Jr, 2012). For the lipid-lipid interactions needed to construct amembrane model, parameters were adapted from a previous MS-CG model developedfor a DOPC:DOPE bilayer in the absence of explicit water molecules (Lu & Voth, 2009).The present work assesses the physical realism that can be obtained by combining the twoprotein and lipid force fields.

The polypeptide is represented as a chain of backbone beads defined at the alpha carbonpositions with up to four CG sites defined within each sidechain as the mass centers ofnon-overlapping subgroups of atoms. In contrast to approximations made in other CGmodels in order to obtain larger time steps, Boltzmann inversion is used to assign accurateharmonic interaction potentials that reproduce the CG bond fluctuations observed inatomistic MD (Hills Jr, 2014). Fourth order polynomials were used to fit the 3-body angledistributions observed for each of the 20 amino acids (Hills Jr, Lu & Voth, 2010). The Cαbackbone is assigned generic but sequence-dependent torsional potentials that allow forthe adoption of either α-helix or β-sheet secondary structure (Karanicolas & Brooks III,2002). The original CgProt parameterization employed a unique site type for the backboneCα bead: ca. To reproduce the bilayer permeation profiles of backbone moieties, version2.4 assigns the existing apolar site type (ap) to Ala residues and the existing polar site (p)type to all other backbone beads (Fig. 1). New site type assignments are explored for select

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 3/22

HNH

CH

CO

3

HNH

CH

CO

2CH2CH2NHCNH2

NH2

Ala* Arg* Asn*

HNH

CH

CO

2CNH2

OAsp*

HNH

CH

CO

2CO

O

HNH

CH

CO

2SH

Cys*

HNH

H

CO

Gly* Gln*

HNH

CH

CO

2CNH2

O

2CH

His

HNH

CH

CO

2

HNN

Ile

HNH

CH

CO

2CH CHCH 33

Leu

HNH

CH

CO

CH

CHCH 3

3

2

Met

HNH

CH

CO

2CH2 CH3S

Phe*

HNH

CH

CO

2

ProHN

CO

HNH

CH

CO

2OH

Ser

Val*

HNH

CH

CO

CH CH33

Thr CH

HNH

CH

CO

3

OH

TrpHNH

CH

CO

2

NH

Tyr*

HNH

CH

CO

2 OH

Lys*

HNH

CH

CO

2CH2CH2CH 32NH

Glu*

HNH

CH

CO

2CO

O

2CH

Figure 1 Site assignments in CgProt 2.4. Alpha carbons not involving alanine are reassigned the polarsite type (sky blue). Other nonbond interaction types include positive (blue), negative (red), and apolar(gray). Asterisks denote new sidechain descriptions for version 2.4.

Full-size DOI: 10.7717/peerj.4230/fig-1

sidechain p or ap sites, but the original nonbond potentials tabulated for each site type paircombination were not modified.

Given the lack of explicit water, dynamics employ the stochastic Langevin dynamicsroutine with a damping friction coefficient of γ = 0.5 ps−1, a significant speedup relativeto protein dynamics in real water (Cerutti et al., 2008). The lateral diffusion of lipids in thebilayer is enhanced 30-fold in CgProt (Fosso-Tande et al., 2017). An effective temperatureof kT = 2.24 kJ/mol is used in the present work as it has been shown to reproducelipid bilayer structure across different liquid phase systems (Hills Jr & McGlinchey, 2016).Nonbond interactions have a 1.2 nm cutoff and the neighbor list is updated every step. Asrecommended by others (Arnarez et al., 2015), the implicit solvent bilayer simulations wereperformed in the fixed volume NVT ensemble. The Gromacs 4.6.7 simulation package wasused (Pall et al., 2015) with a 5 fs integration time step.

Sidechain insertion PMFsCgProt is assessed and refined by comparing the energetics of sidechain bilayer insertion toPMFs from atomistic simulations reported in the literature (De Jong et al., 2013; Johansson& Lindahl, 2008; MacCallum, Bennett & Tieleman, 2008). Umbrella sampling calculationswere performed with a harmonic force constant of 350 kJ mol−1 nm−2 applied between themass centers of the sidechain and lipid bilayer in the z-dimension. As in previous atomisticstudies (MacCallum, Bennett & Tieleman, 2008), an additional copy of the sidechain wasplaced 3.7 nm above the first in order to duplicate the number of data collected in a singlerun. A total of 21 umbrella windows with 0.2 nm spacing were simulated for 50 ns each. Toensure overlap between successive windows, sidechain 1 was restrained at depths spanningz =−3.3 to z =+0.5 nm from the bilayer center, while sidechain 2 spanned z = 0 to 4

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 4/22

nm. The two halves of the PMFs were computed using g_wham (Hub, De Groot & Van derSpoel, 2013) in Gromacs and treated as independent samples to determine the largest errorover the 400 bins. After shifting to zero in bulk water, the maximum PMF error for mostsidechains was between 0.1 and 0.5 kJ/mol at the bilayer center (z = 0). Exceptions wereArg, Lys and Asp, whose strong repulsion at z = 0 had a maximum error of 2.0–2.6 kJ evenafter increasing the simulation length to 250 ns.

A large periodic simulation box (256 DOPC molecules, 9.4 nm box edge) was used tominimize boundary effects in the umbrella simulations (Neale & Pomes, 2016; Nitschke,Atkovska & Hub, 2016). By comparison, a 250 ns simulation performed with a small box(64 DOPC, 4.7×4.7×9 nm) containing two arginine sidechains reveals that the smallsimulation significantly overestimates the steepest part of the PMF at a distance of z = 1 nmfrom the bilayer center. The larger box also decreases the likelihood that the two sidechaincopies influence each other. A comparison PMF obtained from a simulation with onlyone arginine sidechain in a 9.4 nm box fell within the 2.6 kJ margin of error of the twosidechain simulation.

Peptide insertion simulationsThe membrane binding or insertion behavior of synthetic peptides in the DOPC bilayerwas used to test the balance of membrane-protein interactions in CgProt, as has beenperformed with other force fields (Bereau et al., 2015; Bereau & Kremer, 2016; Bereau,Wang & Deserno, 2014; Bond et al., 2007; Hall, Chetwynd & Sansom, 2011; Kar et al., 2014;Pulawski et al., 2016; Ward, Nangia & May, 2017). The sequence GWW(LA)8LWWA(WALP23) of the WALP n series of single-pass transmembrane peptides was simulatedsince its 25.5 Å hydrophobic length (Kim & Im, 2010) nearly matches the hydrophobicthickness of the DOPC bilayer: 29 Å as measured by the mean distance between ester sites.The highly charged transmembrane peptides K2L24K2 and K2(LA)12K2, referred to as L24and (LA)12, respectively, were also assessed for their bilayer insertion stability (Liu et al.,2004; Zhang et al., 1995). Last, the designed amphipathic peptide (LSSLLSL)3, referred toas LS3, was tested for its binding at the membrane-water interface (Lear, Wasserman &DeGrado, 1988).

In versions 2.3 and earlier CgProt relied on terminal charges to anchor peptides withinthe membrane (Fosso-Tande et al., 2017). In this work, each N- and C-terminus is keptneutral in order to rigorously assess the insertion behavior of the current force field and itsnew backbone description. This is consistent with experimental studies (Holt et al., 2009;Zhang et al., 1995) of the designed peptides in which the N- and C-termini are capped withacetyl and amide groups, respectively. Each of eight simulation runs were conducted for 2µs in a 9.5×9.5×10 nm box containing 256 DOPC molecules, with the peptide startingin either a membrane-bound or completely desorbed state. Peptides were restrained inan ideal right-handed α–helix conformation by applying a harmonic GROMOS-96 angle(θ0= 89◦, kθ = 2,000 kJmol−1) between each successive Cα-Cα-Cα and a harmonic torsion(ϕ0 = 50◦, kϕ = 2,000 kJ mol−1 rad−2) for Cα-Cα-C α-Cα terms. The angle of peptideinsertion with respect to the bilayer normal was calculated using the vector spanning alphacarbons 5 and 16 for LS3, 5 and 19 for WALP23, and 8 and 26 for L24/(LA)12.

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 5/22

Peptide insertion PMFsThe PMFs for the insertion of LS3 and WALP23 peptides in a DOPC bilayer weredetermined using umbrella sampling simulations (Bereau et al., 2015; Bereau & Kremer,2016;Ward, Nangia & May, 2017). Umbrella windows incorporated a single peptide placedat 0.2 nm intervals along the z-axis. Starting structures were generated by pulling the peptideinto the bilayer over a 4.6 ns simulation (2×104 kJ mol−1 nm−2 force constant) startingfrom either side of the bilayer at z =±4.6 nm. For each peptide, a total of forty-eightwindows were simulated for 100 ns each using a force constant of 650 kJ mol−1 nm−2

force constant for the peptide center of mass relative to the bilayer normal. Snapshots wererecorded every 1 ps. The first 16 ns of simulation time was excluded from the WHAMcalculation to allow for the peptide orientation to equilibrate.

RESULTS AND DISCUSSIONSidechain permeation profilesThe DOPC bilayer can be divided into four regions (Fig. 2) to aid in comparing to previouswork (Genheden & Essex, 2015; MacCallum, Bennett & Tieleman, 2008). The tail region,found within 1 nm of the bilayer center, contains only hydrophobic tails. The ester regionfrom z = 1 to 1.8 nm begins at the minimum depth of the carbonyls, has a falling lipidtail density, and contains most of the glycerol density. The head group region from 1.8 to2.8 nm contains the largest portion of the phosphate and choline head group density. Thefourth region represents the pure water phase. These trends in DOPC lipid partial densitiesare in agreement with atomistic MD simulations (MacCallum, Bennett & Tieleman, 2008).

PMF insertion profiles are reported for each sidechain from z = 0 to 3.7 nm (Fig. 3).For sidechains in the original force field that did not exhibit expected behavior, alternateCG descriptions were explored. Note that Cys and Pro sidechains consisting of a loneapolar bead are indistinguishable in absence of the Cα site, as are polar Ser and Thr. TheCβ-Cγ virtual bond between apolar CG sites is nearly identical in Ile and Met and leads toPMF profiles that are indistinguishable. The single polar site for Ser/Thr has a 4.4 kJ/molattractive well at z = 1.7 nm and a 10 kJ barrier to crossing the bilayer center at z = 0. Thiscompares well to the atomistic PMFs obtained in OPLS for Ser and Thr, which have 1–4 kJattraction at z = 1.5 nm and a repulsive barrier for z < 1 nm that reaches 15 kJ (MacCallum,Bennett & Tieleman, 2008). For Cys, the single apolar site has an 8.5 kJ repulsion at z = 2.1nm and is attractive by 4–6.6 kJ for z < 1 nm. This corresponds well with the atomisticPMF reported by MacCallum et al. for cysteine: 3 kJ repulsion at z = 2.1 nm and 4–7 kJattraction for z < 1.5 nm. In contrast, the orginal CgProt force field (v.1) employed a polarsite for cysteine, overestimating the energy of insertion into the tail region by 13 kJ/mol.For Val, a second apolar site was added to its CG description to capture the hydrophobicnature of the atomistic PMF for valine: 5 kJ repulsion at z = 2 nm and >10 kJ attractionfor z < 1.5 nm.

Ile and Leu also contain two apolar sites, having a 17–22 kJ attraction for the tail regionof the bilayer. The short 1.6 Å Cβ–Cγ bond of Leu and Val allows them to pack morefavorably inside the tail region than Met and Ile (2.5 Å bond). The PMFs are in qualitative

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 6/22

headestertail water

0

0.5

1

0 1 2 3

dens

ity (g

/mL)

z (nm)

TotalCholine

PhosphateGlycerol

EstersTails

Figure 2 Partial lipid density profiles relative to the bilayer center (z = 0) for the CGDOPC simula-tion. Vertical gray bars divide the bilayer into four regions: hydrophobic tails, esters, lipid head groups,and solvent.

Full-size DOI: 10.7717/peerj.4230/fig-2

agreement with the 15–22 kJ attraction for z < 1 nm in the Ile and Leu atomistic profiles(MacCallum, Bennett & Tieleman, 2008), but there is a larger repulsion in the head groupregion. The three apolar sites in the original description for Phe overestimate its atomisticPMF, which bears a 15 kJ attraction for z < 1.5 nm (De Jong et al., 2013; MacCallum,Bennett & Tieleman, 2008). Better agreement for aromatic phenylalanine was found byassigning the polar site type to Cδ. Neutral histidine contains one apolar and two polarsites in CgProt with a 14 kJ attraction at z = 1.5 nm and 8 kJ repulsion at z = 0. Theatomistic PMF for His is similar in these features but has a smaller energy barrier in thevicinity of z = 2.5 nm (De Jong et al., 2013). Using three polar sites for His reduces thebarrier but results in too strong an attraction for the head group region.

The atomistic PMFs for Trp and Tyr are 21 and 13 kJ attractive at z = 1.3 nm and riseto −5 and +7 kJ at z = 0, respectively (MacCallum, Bennett & Tieleman, 2008). The bestagreement for tryptophan in CgProt is to assign three apolar sidechain sites along with onepolar site at Cγ. For tyrosine, assignment of polar sites at Cγ and Cε in the current versionimproves the permeation profile in the bilayer tail region. A significant energy barrier isfound in both potentials at z = 2.5 nm. As this falls outside the lipid density of the bilayer(Fig. 2), the barrier is likely to affect the kinetics of insertion but not the thermodynamicstabilization for proteins residing in the membrane.

The atomistic PMFs for Gln and Asn are 9 and 7 kJ attractive at z = 1.5 nm and 20 and24 kJ repulsive at z = 0, respectively (De Jong et al., 2013;MacCallum, Bennett & Tieleman,2008). To reproduce this behavior, the Cβ site type of glutamine was changed from apolar topolar, resulting in stronger attraction at the membrane interface and stronger repulsion in

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 7/22

Atomistic MDCgProt v.1CgProt v.2.4

−40

0

40

80

0 1 2 3

Argw

p.p.posap−pos−p

p−p−pos

0 1 2 3

Lyswo

p.p.posap−posp−pos

0 1 2 3

Gluw

ap−negp−neg

0 1 2 3

Aspw

ap−negp−neg

−30

−15

0

15

Ser/Thr

c

ww

p

Gln

c

o

ap−pp−p

Asn

c

w

p−pap−p

p

Cys

ww

pap

−30

−15

0

15

Trp

ap−p−ap−pap−p−ap−ap

Tyr

c

wog

ap−ap−ap−pap−p−ap−p

Phec w

ap−ap−apap−ap−p

His

c

wo

ap−ap−pp−p−p

ap−p−p

−30

−15

0

15

30

Met/Ilec

c

ww

ap−ap

Leuc

wog

ap−ap

Valc

wog

apap−ap

Pro

ww

g

ap

+ +

w w

oww

wwo

ww

c

owooww

wowo

c

w

w wwwo

c

c

**

*

*

Fre

e En

ergy

(kJ/

mol

)

z (nm)

− −

Figure 3 PMF insertion profiles for mass centers of sidechain analogs relative to the bilayer center.PMFs are set to zero in water. PMFs from atomistic simulations are compared in black (MacCallum,Bennett & Tieleman, 2008). The bonding of CG sites is depicted for polar (p), apolar (ap), positive (pos)and negative (neg) types. Shorter bonds (.) were tested for Lys/Arg. Alternate sidechain descriptionsshown in gray are not recommended. Various experimental data are plotted relative to zero for glycine.For sidechain analogs, the water-cyclohexane transfer energy is marked c at z = 0 and the water-octanolenergy is marked o at z = 1.3 nm (Radzicka & Wolfenden, 1988).1Gcor for the octanol solvation of asidechain in a peptide occluded by moderate-sized neighbors is marked w at z = 1.3 nm (Wimley, Creamer& White, 1996).1GGXG for a fully exposed sidechain in GGXGG peptide is marked *g at 1.3 nm for Leu,Val, Pro and Tyr. The sidechain solvation energy for a WLXLL peptide at the POPC interface is marked wat z = 1.7 nm (Wimley & White, 1996).

Full-size DOI: 10.7717/peerj.4230/fig-3

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 8/22

the hydrophobic tail region. For asparagine, the best agreement was obtained by removingits apolar site altogether, which involved increasing the Cα–Cβ bond from 2.0 to 2.6 Å.Glu and Asp are the only sidechains with a monotonic repulsion observed in atomisticMD for all regions in the bilayer, along with an 80 kJ barrier to crossing the bilayer in theircharged form. CgProt reproduces the strong repulsion in the tail and ester regions usingone polar and one negative site type. In place of a screened Coulomb potential, the forcefield energy function successfully combines electrostatic and van der Waals forces into asingle tabulated function for each of the positive and negative site types (Lu & Voth, 2009;Noid et al., 2008).

The atomistic PMFs for Arg+ and Lys+ are +50 kJ repulsive at z = 0 but have a largeattractive well spanning z = 0.8 to 2.5 nm with a minimum of −20 kJ/mol (De Jong et al.,2013; MacCallum, Bennett & Tieleman, 2008). Arginine and lysine were too repulsive forthe membrane in CgProt v.1. The solution employed in CgProt 2.4 is to reassign theirapolar Cβ sites to be polar. An alternate sidechain description was also tested consisting oftwo polar sites and a positive site connected by short 1.5 Å bonds (p.p.pos). Experimentalmeasures of hydrophobicity were used to discriminate between alternate CG descriptions,which are discussed in the next section.

There has been much interest in the attraction of positive-charged sidechains forthe lipid bilayer (Gumbart & Roux, 2012; MacCallum & Tieleman, 2011; Sun, Forsman &Woodward, 2015). Findings are now emerging as to its potential functional relevance forpeptides (Gleason et al., 2013; Hu, Sinha & Patel, 2014; MacCallum, Bennett & Tieleman,2011; Nakao et al., 2016; Rice & Wereszczynski, 2017). The favorable attraction of basicresidues for the ester bilayer region has been dubbed snorkeling. A bilayer defect is createdin which polar lipid head groups dive down to solvate the positive charge. A representativesnapshot from the CgProt umbrella simulations (Fig. 4) shows the arginine sidechainnear the center of the bilayer producing local changes in the depth of glycerol, phosphateand choline groups, similar to results observed in atomistic MD (MacCallum, Bennett &Tieleman, 2008).

Experimental partition energiesExperimental values for the partitioning of sidechain analog compounds from waterto cyclohexane were correlated with the free energy of transfer to the bilayer center atz = 0 (MacCallum, Bennett & Tieleman, 2008; Pogorelov et al., 2014; Vorobyov et al., 2016)computed from the CgProt 2.4 PMFs. A linear regression was performed for all sidechainsexcept charged Asp, Glu and Lys. Since the experiments did not control the ionizationstate (Radzicka & Wolfenden, 1988), the Asp, Glu and Lys sidechains are expected to beneutral in the nonpolar environment of cyclohexane based on analysis of their pKa valuesand solvation energies from atomistic MD (MacCallum, Bennett & Tieleman, 2008). Astrong correlation (Pearson coefficient of r = 0.95) is observed between CgProt 2.4 andexperiment (Fig. 5A), similar to the correlation between cyclohexane solvation energiesand the atomistic PMF values of MacCallum et al. (Table 1).

The transfer free energy of sidechain analogs from water to octanol was measured in thesame study (Radzicka & Wolfenden, 1988). Positing that octanol has properties analogous

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 9/22

Figure 4 Snapshot of p-p-pos Arg (cyan-cyan-blue spheres) restrained near the center of the bilayer. Abilayer defect solvates the CgProt sidechain similar to observations in atomistic MD (MacCallum, Bennett& Tieleman, 2008). DOPC head groups are colored as blue (choline), red (phosphate) and green (glycerol)balls. Lipid tails are traced as sticks.

Full-size DOI: 10.7717/peerj.4230/fig-4

Table 1 MacCallum et al. and CG PMFs versus experiment.

x: Experiment (1G) y: PMF Pearsoncorrelation, r

Slope Intercept(kJ/mol)

Cyclohexane-water Atomistica 0.98 0.89 1.3Cyclohexane-water CgProta 0.95 0.88 −1.2Octanol-water Atomisticb 0.92c 1.6 −2.6Octanol-water CgProtb 0.78 1.1 −3.7W-W octanol (1Gcor) Atomisticb 0.95c 1.6 0.8W-W octanol (1Gcor) CgProtb 0.91 2.0 5.3W-W POPC interface Atomisticd 0.86c 1.6 −1.6W-W POPC interface CgProtd 0.51e 0.68 −0.4

Notes.az = 0 nm.bz = 1.3 nm.cExcludes outliers Arg and Lys.dz = 1.7 nm.eExcludes outliers His, Arg and Gln.

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 10/22

−40

0

40

80

−40 0 40 80

y = 0.88x − 1.2r = 0.95

A

CgP

rot (

kJ/m

ol)

cyclohexane−water

NC

Q

LMF

ST

WY

V

R

I

H

−20

−10

0

10

−10 0 10

y = 1.11x − 3.7r = 0.78

B

CgP

rot (

kJ/m

ol)

octanol−water

N

Q

L

MF

ST

W YV

I

R

H

−20

0

20

40

−20 −10 0 10 20

y = 1.98x + 5.3r = 0.91

C

CgP

rot (

kJ/m

ol)

W−W octanol

NCQ

LMF

ST

W YVI

R+

K+

D−E−

H

P

−10

0

10

−10 0 10

y = 0.68x − 0.4r = 0.51

D

CgP

rot (

kJ/m

ol)

POPC−interface

N

D−C

L

M

FSTW

Y

V

E−I

P

K+

R*H*Q*

(Trp)(Met)

(Ser)(Asn)

(Thr/Ser)(Cys/Pro)

(Ile) (Met)

(Ser/Thr)

(Asp/Glu)

(Val)

Figure 5 Linear regression of CgProt 2.4 sidechain solvation energies with experimental data. Sin-gle letter amino acid codes are plotted along with select 3-letter codes in parentheses. Water to cyclohex-ane (A) and water to octanol (B) transfer energies of unionized sidechain analogs (Radzicka & Wolfenden,1988) are correlated to PMF values at z = 0 and z = 1.3 nm, respectively. Wimley-White peptide residuetransfer energies relative to glycine are correlated to PMF values at z = 1.3 and 1.7 nm for octanol-water(C) and the POPC interface (D), respectively.

Full-size DOI: 10.7717/peerj.4230/fig-5

to the ester region of the bilayer, we correlated the octanol solvation energies to the PMFvalues of MacCallum et al. for transferring solute to a membrane depth of z = 1.3 nm andobtained a good correlation of r = 0.92 upon exclusion of outliers Arg and Lys (Table 1).For the PMFs calculated in CgProt 2.4, a correlation of r = 0.78 was found upon exclusionof sidechains Asp, Glu and Lys (Fig. 5B). Arginine, which is predicted to be protonated inthe bilayer ester region (MacCallum, Bennett & Tieleman, 2008), falls on the regression line.

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 11/22

For a measure of sidechains occluded by neighboring residues of moderate size (Wimley,Creamer & White, 1996), the octanol-water partitioning of WLXLL peptides relative toglycine (1Gcor) was correlated with the PMF values at z = 1.3 nm. A strong correlation(r = 0.95) is observed for the atomistic PMFs, excluding outliers Arg and Lys. Similarly, aprominent linear correlation (r = 0.91) is observed for all amino acid sidechains in CgProt(Fig. 5C). The experiments of Wimley et al. controlled for protonation state, allowing forcomparison to all charged sidechain simulations. The CgProt PMFs also correlate well with1GGXG values for a fully exposed residue in a glycine peptide (r = 0.91).

To compare the partitioning of sidechains to the polar environment of the bilayerinterface, we take the whole-residue solvation energies derived from the WLXLL peptideseries in POPC and subtract the +0.04 kJ/mol value for glycine (Wimley & White, 1996).Using the free energy at a depth of z = 1.7 nm, a good correlation of r = 0.86 is obtainedfor the atomistic PMFs of MacCallum et al., excluding outliers Arg and Lys. For CgProt,a weak correlation of r = 0.51 is observed. His, Arg and Gln were found to be outliersand were excluded from the linear regression (Fig. 5D). The various transfer energiesfor cyclohexane, octanol, and POPC measurements are plotted at z = 0, 1.3, and 1.7nm, respectively, in Fig. 3. Together the experimental data points serve as a guide forselecting between alternate CG descriptions. This is particularly useful for Arg and Lys,for which more positive solvation energies are measured in experiments as compared toatomistic MD.

Peptide backbone parametersEarlier versions of CgProt enabled the rapid insertion of peptides with charged N- andC-termini in unbiased MD simulations (Fosso-Tande et al., 2017). This was due to theattractive nature of the unique backbone type (ca) for the bilayer core. PMFs are calculatedfor the insertion of one and two-residue ca backbone units in the bilayer, revealing afree energy minimum at z = 0.7 nm of −16.5 and −35 kJ/mol, respectively (Fig. 6).This is in stark contrast with the +1.15 kcal/mol water to octanol transfer energy for theglycyl -CH2CONH- unit (Wimley, Creamer & White, 1996). PMFs calculated for one andtwo-residue backbone units using the polar site type (p) are in better agreement withPMFs obtained from atomistic MD for acetamide and other backbone mimics (Bereau& Kremer, 2016; Sandoval-Perez, Pluhackova & Bockmann, 2017), as well as the positivepartition energies obtained from experiment. For the CgProt representation of alanine,the backbone bead is the sole CG site. The PMF of methane (CH4) has a −8.4 kJ/mol freeenergy minimum at z = 0 and its cyclohexane partition energy is−7.6 kJ/mol (Radzicka &Wolfenden, 1988). Alanine residues are therefore represented using the apolar site type (ap),which provides good agreement with the Wimley-White whole-residue values (Fig. 6).

Peptide insertion propertiesSimulations of peptide insertion were used to test the balance of backbone and sidechaininteractions in CgProt 2.4 (Bereau & Kremer, 2016). Multiple independent simulationswere performed for the designed helical peptides LS3, WALP23, L24 and (LA)12 startingfrom either membrane-bound or desorbed states. In each simulation, the peptide adopted a

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 12/22

−40

−20

0

20

40

0 1 2 3

Cα

c

ww

PMF

(kJ/

mol

)

v.1 (ca)v2.4 (p)

0 1 2 3

Cα−Cα

d2w2w

z (nm)

v.1 (ca−ca) v2.4 (p−p)

0 1 2 3

Ala

w

v.1 (ca)v2.4 (ap)

o w*

Figure 6 CgProt PMFs for one and two-residue backbone moieties. CgProt v.1 employed a unique al-pha carbon site type (ca) as opposed to polar sites (p). Cα: The atomistic PMF for acetamide is drawnin black as an approximation for the glycyl -CH2CONH- unit (MacCallum, Bennett & Tieleman, 2008).Cyclohexane and octanol measurements for acetamide are labeled c and o at 0 and 1.3 nm, respectively.Whole-residue Wimley-White measurements of glycine are labeled w at 1.3 and 1.7 nm. Cα–Cα: Twicethe atomistic PMF for acetamide is shown (black line). 2w corresponds to twice the Wimley-White valuefor glycine at 1.3 and 1.7 nm. The water-octanol transfer energy of the Gly dipeptide is labeled d* at 1.3nm (Fauchere & Pliska, 1983). Ala: The atomistic PMF of methane (CH4) is shifted to account for glycine’s1.15 kcal octanol solvation penalty (black). Whole-residue Wimley-White values for alanine are labeled w.The PMF of a single CgProt apolar site (ap) is shown in aqua.

Full-size DOI: 10.7717/peerj.4230/fig-6

Table 2 Peptide orientation during 1–2 µs of simulation time.

Peptide Starting state Helix tilta (◦) zcenterb (Å) zN-termc (Å) zC-term

c (Å)

LS3 desorbed 85.1± 5.5 12.4± 1.1 11.4± 1.7 13.4± 1.8LS3 interfacial 84.6± 5.4 12.2± 1.2 11.0± 1.7 13.4± 1.8L24 desorbed 85.4± 3.7 19.6± 1.5 19.0± 1.5 18.4± 1.5L24 transmembrane 32.5± 6.2 −0.1± 0.2 16.8± 1.8 −18.2± 2.0(LA)12 desorbed 88.9± 3.8 13.6± 1.1 14.1± 1.4 13.8± 1.5(LA)12 transmembrane 35.4± 5.4 0.7± 1.3 17.5± 1.4 −17.0± 1.7WALP23 desorbed – – – –WALP23 transmembrane 41.0± 7.3 0.9± 1.5 13.3± 1.8 −10.2± 2.5

Notes.aMean angle± standard deviation of helix axis with respect to bilayer normal.bMean displacement of helix center of mass from bilayer center.cMean distance of terminal Cα from bilayer center.

single stable orientation within themembrane, with the exception of the desorbedWALP23simulation, in which the peptide did not adsorb onto the membrane. The kinetic barrierto WALP23 insertion can be explained by the barrier in the PMF for tryptophan at themembrane interface near z = 2.5 nm (Fig. 3). The orientation of the helix axis relative to thebilayer normal is shown for each simulation in Table 2, along with the membrane depth ofthe peptide center of mass. The final equilibrated conformation for each membrane-boundsimulation is shown in Fig. 7.

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 13/22

Figure 7 Equilibrated peptide orientations inµs membrane-bound simulations. The LS3 peptide in-serts in an interfacial orientation (A), while L24, (LA)12, and WALP23 remain in a transmembrane orienta-tion (B, C & D). The surface of lipid head groups is shown in transparent blue (choline), red (phosphate),green (glycerol) and white (esters). The Cα backbone is traced in cyan with neutral N- and C-termini la-beled blue and red, respectively. Spheres are colored for polar sites belonging to serine (gold) and trypto-phan (green) sidechains and for the positive charge of lysine (cream).

Full-size DOI: 10.7717/peerj.4230/fig-7

Interfacial conformations were observed in both unrestrained LS3 simulations that areindistinguishable from each other. The N-terminus sits 2 Å lower in themembrane than theC-terminus (Table 2). Given that the termini were uncharged this is likely a property of thegeometry of the peptide. In contrast to CgProt 2.3 simulations with charged termini and thehydrophobic ca type backbone in which LS3 preferred the transmembrane configuration(Fosso-Tande et al., 2017), the umbrella sampling PMF obtained for insertion of the neutral

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 14/22

−200−160

−120

−80

−40

0 40

0 1 2 3 4

LS3PMF

(kJ/

mol

)

z (nm)

−100−80

−60

−40

−20

0 20

0 1 2 3 4

WALP23

z (nm)

A B

Figure 8 Peptide insertion PMFs. Independent PMFs were obtained for the permeation of LS3 (A) andWALP23 (B) in both leaflets of the bilayer (dashed lines). The symmetrized PMF (solid line) reveals aglobal energy minimum for both interfacial LS3 and transmembrane WALP23.

Full-size DOI: 10.7717/peerj.4230/fig-8

LS3 confirms that the interfacial conformation is the only preferred energy minimum inCgProt 2.4 (Fig. 8A). The PMF at z = 0 is 65 kJ/mol higher in energy than the interfacial statefound at a depth of z = 1.2 nm. When restrained to the center of the bilayer, LS3 adopts anunstable transmembrane orientation in which the serine sidechains pull lipid head groupsinto the bilayer interior. While phospholipid flip-flop has not been observed in puremembrane simulations with CgProt, enhancement of lipid flip-flop by transmembranehydrophilic residues is consistent with recent experiments on endoplasmic reticulumpeptide sequences (Nakao et al., 2016). The pronounced stabilization of interfacial LS3 isin contrast to unrestrained simulations with the Bond CG model (Bond, Wee & Sansom,2008; Hall, Chetwynd & Sansom, 2011), a variant of Martini, in which the interfacial andtransmembrane conformations were observed with 80% and 20% frequency, respectively.In situ, the amphipathic LS3 peptide is only expected to adopt a transmembrane orientationwhen residing in the membrane as part of a homo-oligomeric bundle, which serves as anion channel (Nguyen, Liu & Moore, 2013).

The PMF for the bilayer insertion of WALP23 reveals that the transmembraneconfiguration centered near z = 0 is preferred over the interfacial state in the vicinityof z = 0.9 nm by 15.5 kJ/mol (Fig. 8B). The interfacial state is a local energy minimum buthas a minimal 1 kJ barrier separating it from the transmembrane state. It is therefore notexpected to be stable in unbiased molecular dynamics simulations. The stable insertion ofneutral WALP is a considerable improvement in CgProt 2.4. The 24±5

◦

tilt angle obtainedby fluorescence spectroscopy (Holt et al., 2009) is within two standard deviations of therange observed in the unrestrained simulation (41±7

◦

). In contrast, the hydrophobic catype backbone employed in previous versions causes uncharged WALP peptide termini tosubmerge in the hydrophobic tail region of the bilayer (Fig. S1). The 93 kJ/mol stabilizationof transmembrane WALP in the current version is comparable to the 146 kJ/mol energy

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 15/22

minimum observed for the PLUM force field in a POPC bilayer (Bereau et al., 2015). Incontrast, transmembrane and interfacial WALP orientations have been observed with 80%and 20% frequency, respectively, in variants of the Martini model (Bereau & Kremer, 2016;Hall, Chetwynd & Sansom, 2011).

The previous ca type hydrophobic backbone enabled the insertion of peptides tobe observed within 50 ns of unbiased molecular dynamics simulation. Optimizing thebalance of backbone and sidechain energetics in the current version, however, resultedin no membrane insertion events being observed in 8 µs of simulations starting from thedesorbed state. This includes the L24 and (LA)12 peptides, which were observed to adsorbonto the membrane surface but not cross over to the other monolayer on the timescaleof the simulation. This can be explained by the presence of a 10 kJ barrier for the polartype backbone sites of non-alanine residues to cross the center of the bilayer (Fig. 6).Additionally, the terminal lysines in the peptides each have a 4 kJ barrier to penetrating thebilayer interface. These studies highlight the need for either manual insertion of a proteininto the bilayer or the use of biased sampling methods.

CONCLUSIONThe CgProt nonbond parameters have been refined to improve the permeation behaviorfor the peptide backbone and amino acid sidechain analogs (Hu, Sinha & Patel, 2014). Thebackbone unit has beenmade less attractive for the bilayer by assigning the polar site type fornon-alanine residues and the apolar site type for alanine. Additional polar sites have beenassigned for sidechains Arg, Asp, Lys, Gln, Glu, Tyr, and Phe. The permeation of version2.4 sidechains correlates well with PMFs from atomistic simulation and experimentalhydrophobicity scales based on cyclohexane-to-water and octanol-to-water partitionenergies. Atomistic simulations with OPLS overestimate the attraction of Gln, Arg and Lysfor the ester region of the bilayer. The permeation of Asn and Lys is much improved inCgProt 2.4 compared to v.1, while Gln and Arg experience a favorable attraction similarto atomistic simulations. That Arg is more attractive to the membrane than Lys in CgProtis consistent with the enhanced activity of arginine containing antimicrobial peptidesversus their lysine containing counterparts (Rice & Wereszczynski, 2017). Note that threesidechain assignments employed in v.2.3 are not recommended based on their permeationbehavior: the p.p.pos description of Lys, the ap-p-ap-p description of Trp, and the p-pdescription of Asn. Previous simulations with v.2.3 revealed that the positive N-terminushas enhanced permeation properties compared to the negative C-terminus (Fosso-Tande etal., 2017). The enhanced membrane permeation of positive charges as opposed to negativesite types is supported by the CgProt PMF calculations.

Refining the CgProt force field by matching sidechain permeation profiles resulted inimproved simulations of membrane-associated peptides. While previous CgProt versionsrelied on the use of terminal charges to anchor peptides and proteins in the membrane(Fosso-Tande et al., 2017), the global energy minimum is now correctly obtained for theinterfacial LS3 and transmembraneWALP peptides. The improved backbone and sidechaindescriptions in CgProt 2.4 enable the simulation of diverse membrane protein systems

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 16/22

and will serve to guide future model development efforts. The CgProt force field is aneffective tool for studies of lipid-protein interactions and protein conformational change.To date protein-protein association has not been studied in CgProt. Refinements inamino acid hydrophobicity are likely key to reproducing experimentally observed dimerstructures, which has proved a considerable challenge in CG models such as Martini(Javanainen, Martinez-Seara & Vattulainen, 2017). The CgProt tools and parametersneeded for implementing molecular dynamics simulations in the Gromacs softwarepackage are included in the Supplemental Information, along with equilibrated bilayercoordinates.

ACKNOWLEDGEMENTSR.D.H. thanks Jacob Fosso-Tande for valuable assistance.

ADDITIONAL INFORMATION AND DECLARATIONS

FundingThis work was supported by a National Science Foundation grant (MCB-1516826). Thefunders had no role in study design, data collection and analysis, decision to publish, orpreparation of the manuscript.

Grant DisclosuresThe following grant information was disclosed by the author:National Science Foundation: MCB-1516826.

Competing InterestsThe authors declare there are no competing interests.

Author Contributions• Ronald D. Hills, Jr conceived and designed the experiments, performed the experiments,analyzed the data, contributed reagents/materials/analysis tools, wrote the paper,prepared figures and/or tables, reviewed drafts of the paper.

Supplemental InformationSupplemental information for this article can be found online at http://dx.doi.org/10.7717/peerj.4230#supplemental-information.

REFERENCESArnarez C, Uusitalo JJ, MasmanMF, Ingolfsson HI, De Jong DH,MeloMN, Periole

X, De Vries AH, Marrink SJ. 2015. Dry Martini, a coarse-grained force field forlipid membrane simulations with implicit solvent. Journal of Chemical Theory andComputation 11:260–275 DOI 10.1021/ct500477k.

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 17/22

Bennett WF, Tieleman DP. 2013. Computer simulations of lipid membranedomains. Biochimica et Biophysica Acta—Biomembranes 1828:1765–1776DOI 10.1016/j.bbamem.2013.03.004.

Bereau T, Bennett WF, Pfaendtner J, DesernoM, KarttunenM. 2015. Folding andinsertion thermodynamics of the transmembrane WALP peptide. Journal of ChemicalPhysics 143:Article 243127 DOI 10.1063/1.4935487.

Bereau T, Kremer K. 2016. Protein-backbone thermodynamics across the membrane in-terface. Journal of Physical Chemistry B 120:6391–6400 DOI 10.1021/acs.jpcb.6b03682.

Bereau T,Wang ZJ, DesernoM. 2014.More than the sum of its parts: coarse-grainedpeptide-lipid interactions from a simple cross-parametrization. Journal of ChemicalPhysics 140:Article 115101 DOI 10.1063/1.4867465.

Bond PJ, Holyoake J, Ivetac A, Khalid S, SansomMS. 2007. Coarse-grained moleculardynamics simulations of membrane proteins and peptides. Journal of StructuralBiology 157:593–605 DOI 10.1016/j.jsb.2006.10.004.

Bond PJ, Wee CL, SansomMS. 2008. Coarse-grained molecular dynamics simulationsof the energetics of helix insertion into a lipid bilayer. Biochemistry 47:11321–11331DOI 10.1021/bi800642m.

Cerutti DS, Duke R, Freddolino PL, Fan H, Lybrand TP. 2008. Vulnerability in popularmolecular dynamics packages concerning Langevin and Andersen dynamics. Journalof Chemical Theory and Computation 4:1669–1680 DOI 10.1021/ct8002173.

De Jong DH, Singh G, Bennett WFD, Arnarez C,Wassenaar TA, Schafer LV, PerioleX, Tieleman DP, Marrink SJ. 2013. Improved parameters for the Martini coarse-grained protein force field. Journal of Chemical Theory and Computation 9:687–697DOI 10.1021/ct300646g.

Fauchere J, Pliska V. 1983.Hydrophobic parameters II of amino-acid side chains fromthe partitioning of N-acetyl-amino-acid amides. European Journal of MedicinalChemistry—Chimica Therapeutica 18:369–375.

Fosso-Tande J, Black C, Aller SG, Lu L, Hills Jr RD. 2017. Simulation of lipid-proteininteractions with the CgProt force field. AIMS Molecular Science 4:352–369DOI 10.3934/molsci.2017.3.352.

Ganesan SJ, Matysiak S. 2014. Role of backbone dipole interactions in the formation ofsecondary and supersecondary structures of proteins. Journal of Chemical Theory andComputation 10:2569–2576 DOI 10.1021/ct401087a.

Genheden S, Essex JW. 2015. A simple and transferable all-atom/coarse-grained hybridmodel to study membrane processes. Journal of Chemical Theory and Computation11:4749–4759 DOI 10.1021/acs.jctc.5b00469.

Gleason NJ, Vostrikov VV, Greathouse DV, Koeppe II RE. 2013. Buried lysine, but notarginine, titrates and alters transmembrane helix tilt. Proceedings of the NationalAcademy of Sciences of the United States of America 110:1692–1695DOI 10.1073/pnas.1215400110.

Gumbart J, Roux B. 2012. Determination of membrane-insertion free energies bymolecular dynamics simulations. Biophysical Journal 102:795–801DOI 10.1016/j.bpj.2012.01.021.

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 18/22

Hall BA, Chetwynd AP, SansomMS. 2011. Exploring peptide-membrane interac-tions with coarse-grained MD simulations. Biophysical Journal 100:1940–1948DOI 10.1016/j.bpj.2011.02.041.

HanW,Wan CK,Wu YD. 2008. Toward a coarse-grained protein model coupled witha coarse-grained solvent model: solvation free energies of amino acid side chains.Journal of Chemical Theory and Computation 4:1891–1901 DOI 10.1021/ct800184c.

Hedger G, SansomMS. 2016. Lipid interaction sites on channels, transporters and recep-tors: recent insights from molecular dynamics simulations. Biochimica et BiophysicaActa—Biomembranes 1858:2390–2400 DOI 10.1016/j.bbamem.2016.02.037.

Hills Jr RD. 2014. Balancing bond, nonbond, and Go-like terms in coarse grain simu-lations of conformational dynamics.Methods in Molecular Biology 1084:123–140DOI 10.1007/978-1-62703-658-0_7.

Hills Jr RD, Lu L, Voth GA. 2010.Multiscale coarse-graining of the protein energy land-scape. PLOS Computational Biology 6:e1000827 DOI 10.1371/journal.pcbi.1000827.

Hills Jr RD, McGlinchey N. 2016.Model parameters for simulation of physiologicallipids. Journal of Computational Chemistry 37:1112–1118 DOI 10.1002/jcc.24324.

Holt A, Koehorst RB, Rutters-Meijneke T, GelbMH, Rijkers DT, HemmingaMA,Killian JA. 2009. Tilt and rotation angles of a transmembrane model peptideas studied by fluorescence spectroscopy. Biophysical Journal 97:2258–2266DOI 10.1016/j.bpj.2009.07.042.

Hu Y, Sinha SK, Patel S. 2014. Reconciling structural and thermodynamic predictionsusing all-atom and coarse-grain force fields: the case of charged oligo-argininetranslocation into DMPC bilayers. Journal of Physical Chemistry B 118:11973–11992DOI 10.1021/jp504853t.

Hub JS, De Groot BL, Van der Spoel D. 2013. g_wham: a free weighted histogramanalysis implementation including robust error and autocorrelation estimates.Journal of Chemical Theory and Computation 6:3713–3720 DOI 10.1021/ct300646g.

JavanainenM,Martinez-Seara H, Vattulainen I. 2017. Excessive aggregation ofmembrane proteins in the Martini model. PLOS ONE 12:e0187936DOI 10.1371/journal.pone.0187936.

Johansson AC, Lindahl E. 2008. Position-resolved free energy of solvation for aminoacids in lipid membranes from molecular dynamics simulations. Proteins: Structure,Function, and Bioinformatics 70:1332–1344 DOI 10.1002/prot.21629.

Kar P, Gopal SM, Cheng YM, Panahi A, Feig M. 2014. Transferring the PRIMO coarse-grained force field to the membrane environment: simulations of membraneproteins and helix-helix association. Journal of Chemical Theory and Computation10:3459–3472 DOI 10.1021/ct500443v.

Karanicolas J, Brooks III CL. 2002. The origins of asymmetry in the folding transitionstates of protein L and protein G. Protein Science 11:2351–2361DOI 10.1110/ps.0205402.

Kim T, ImW. 2010. Revisiting hydrophobic mismatch with free energy simulationstudies of transmembrane helix tilt and rotation. Biophysical Journal 99:175–183DOI 10.1016/j.bpj.2010.04.015.

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 19/22

Kmiecik S, Gront D, Kolinski M,Wieteska L, Dawid AE, Kolinski A. 2016. Coarse-grained protein models and their applications. Chemical Reviews 116:7898–7936DOI 10.1021/acs.chemrev.6b00163.

Lear JD,Wasserman ZR, DeGradoWF. 1988. Synthetic amphiphilic peptide models forprotein ion channels. Science 240:1177–1181 DOI 10.1126/science.2453923.

Liu F, Lewis RN, Hodges RS, McElhaney RN. 2004. Effect of variations in the structureof a polyleucine-based alpha-helical transmembrane peptide on its interactionwith phosphatidylethanolamine bilayers. Biophysical Journal 87:2470–2482DOI 10.1529/biophysj.104.046342.

Lu L, Voth GA. 2009. Systematic coarse-graining of a multicomponent lipid bilayer.Journal of Physical Chemistry B 113:1501–1510 DOI 10.1021/jp809604k.

MaH, Irudayanathan FJ, JiangW, Nangia S. 2015. Simulating gram-negative bac-terial outer membrane: a coarse grain model. Journal of Physical Chemistry B119:14668–14682 DOI 10.1021/acs.jpcb.5b07122.

MacCallum JL, Bennett WF, Tieleman DP. 2008. Distribution of amino acids ina lipid bilayer from computer simulations. Biophysical Journal 94:3393–3404DOI 10.1529/biophysj.107.112805.

MacCallum JL, Bennett WF, Tieleman DP. 2011. Transfer of arginine into lipid bilayersis nonadditive. Biophysical Journal 101:110–117 DOI 10.1016/j.bpj.2011.05.038.

MacCallum JL, Tieleman DP. 2011.Hydrophobicity scales: a thermodynamic lookingglass into lipid-protein interactions. Trends in Biochemical Sciences 36:653–662DOI 10.1016/j.tibs.2011.08.003.

Marrink SJ, Tieleman DP. 2013. Perspective on the Martini model. Chemical SocietyReviews 42:6801–6822 DOI 10.1039/c3cs60093a.

Nakao H, Ikeda K, Ishihama Y, NakanoM. 2016.Membrane-spanning sequences inendoplasmic reticulum proteins promote phospholipid flip-flop. Biophysical Journal110:2689–2697 DOI 10.1016/j.bpj.2016.05.023.

Neale C, Pomes R. 2016. Sampling errors in free energy simulations of small moleculesin lipid bilayers. Biochimica et Biophysica Acta—Biomembranes 1858:2539–2548DOI 10.1016/j.bbamem.2016.03.006.

Nguyen TH, Liu Z, Moore PB. 2013.Molecular dynamics simulations of homo-oligomeric bundles embedded within a lipid bilayer. Biophysical Journal105:1569–1580 DOI 10.1016/j.bpj.2013.07.053.

Nitschke N, Atkovska K, Hub JS. 2016. Accelerating potential of mean force calcu-lations for lipid membrane permeation: system size, reaction coordinate, solute-solute distance, and cutoffs. Journal of Chemical Physics 145:Article 125101DOI 10.1063/1.4963192.

NoidWG, Chu JW, Ayton GS, Krishna V, Izvekov S, Voth GA, Das A, AndersenHC. 2008. The multiscale coarse-graining method. I. A rigorous bridge betweenatomistic and coarse-grained models. Journal of Chemical Physics 128:Article 244114DOI 10.1063/1.2938860.

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 20/22

NoidWG, Chu JW, Ayton GS, Voth GA. 2007.Multiscale coarse-graining and structuralcorrelations: connections to liquid-state theory. Journal of Physical Chemistry B111:4116–4127 DOI 10.1021/jp068549t.

Pall S, AbrahamMJ, Kutzner C, Hess B, Lindahl E. 2015. Tackling exascale softwarechallenges in molecular dynamics simulations with GROMACS. In: Solving SoftwareChallenges for Exascale EASC 2014. Lecture notes in computer science, 8759. 3–27.

Pogorelov TV, Vermaas JV, Arcario MJ, Tajkhorshid E. 2014. Partitioning of aminoacids into a model membrane: capturing the interface. Journal of Physical ChemistryB 118:1481–1492 DOI 10.1021/jp4089113.

Poyry S, Vattulainen I. 2016. Role of charged lipids in membrane structures: in-sight given by simulations. Biochimica et Biophysica Acta—Biomembranes1858:2322–2333 DOI 10.1016/j.bbamem.2016.03.016.

PulawskiW, JamrozM, Kolinski M, Kolinski A, Kmiecik S. 2016. Coarse-grainedsimulations of membrane insertion and folding of small helical proteins usingthe CABS model. Journal of Chemical Information and Modeling 56:2207–2215DOI 10.1021/acs.jcim.6b00350.

Radzicka A,Wolfenden R. 1988. Comparing the polarities of the amino acids: side-chain distribution coefficients between the vapor phase, cyclohexane, 1-ocatnol, andneutral aqueous solution. Biochemistry 27:1664–1670 DOI 10.1021/bi00405a042.

Rice A,Wereszczynski J. 2017. Probing the disparate effects of arginine and lysineresidues on antimicrobial peptide/bilayer association. Biochimica et BiophysicaActa—Biomembranes 1859:1941–1950 DOI 10.1016/j.bbamem.2017.06.002.

Sandoval-Perez A, Pluhackova K, Bockmann RA. 2017. Critical comparison of biomem-brane force fields: protein-lipid interactions at the membrane interface. Journal ofChemical Theory and Computation 13:2310–2321 DOI 10.1021/acs.jctc.7b00001.

Singh G, Tieleman DP. 2011. Using the Wimley-White hydrophobicity scale as a directquantitative test of force fields: the MARTINI coarse-grained model. Journal ofChemical Theory and Computation 7:2316–2324 DOI 10.1021/ct2002623.

Sun D, Forsman J, Woodward CE. 2015. Evaluating force fields for the computationalprediction of ionized arginine and lysine side-chains partitioning into lipid bi-layers and octanol. Journal of Chemical Theory and Computation 11:1775–1791DOI 10.1021/ct501063a.

Van Oosten B, Harroun TA. 2016. AMARTINI extension for Pseudomonas aeruginosaPAO1 lipopolysaccharide. Journal of Molecular Graphics & Modelling 63:125–133DOI 10.1016/j.jmgm.2015.12.002.

Venable RM, Ingolfsson HI, Lerner MG, Perrin Jr BS, Camley BA, Marrink SJ, BrownFLH, Pastor RW. 2017. Lipid and peptide diffusion in bilayers: the Saffman-Delbruck model and periodic boundary conditions. Journal of Physical Chemistry B121:3443–3457 DOI 10.1021/acs.jpcb.6b09111.

Vorobyov I, Kim I, Chu ZT,Warshel A. 2016. Refining the treatment of membraneproteins by coarse-grained models. Proteins: Structure, Function, and Bioinformatics84:92–117 DOI 10.1002/prot.24958.

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 21/22

Ward AB, Guvench O, Hills Jr RD. 2012. Coarse grain lipid-protein molecular in-teractions and diffusion with MsbA flippase. Proteins: Structure, Function, andBioinformatics 80:2178–2190 DOI 10.1002/prot.24108.

WardMD, Nangia S, May ER. 2017. Evaluation of the hybrid resolution PACE model forthe study of folding, insertion, and pore formation of membrane associated peptides.Journal of Computational Chemistry 38:1462–1471 DOI 10.1002/jcc.24694.

White SH,WimleyWC. 1998.Hydrophobic interactions of peptides with membraneinterfaces. Biochimica et Biophysica Acta 1376:339–352DOI 10.1016/S0304-4157(98)00021-5.

WimleyWC, Creamer TP,White SH. 1996. Solvation energies of amino acid side chainsand backbone in a family of host-guest pentapeptides. Biochemistry 35:5109–5124DOI 10.1021/bi9600153.

WimleyWC,White SH. 1996. Experimentally determined hydrophobicity scalefor proteins at membrane interfaces. Nature Structural Biology 3:842–848DOI 10.1038/nsb1096-842.

Zhang YP, Lewis RN, Henry GD, Sykes BD, Hodges RS, McElhaney RN. 1995.Peptide models of helical hydrophobic transmembrane segments of membraneproteins. 1. Studies of the conformation, intrabilayer orientation, and amidehydrogen exchangeability of Ac-K2-(LA)12-K2-amide. Biochemistry 34:2348–2361DOI 10.1021/bi00007a031.

Hills, Jr (2018), PeerJ, DOI 10.7717/peerj.4230 22/22